dr. edward k.l. chan principal investigator the study on ... file-2 log likelihood statistics for...
TRANSCRIPT
Dr. Edward K.L. ChanPrincipal Investigator
The Study on Child Abuse and Spouse BatteringThe University of Hong Kong
立法會CB(2)2345/06-07)號文件LC Paper No. CB(2) 2345/06-07)
The Study on Child Abuse and Spouse Battering- Part Two
Objective: To develop and validate assessment tools to facilitate early identification of cases at risk of child abuse and spouse battering and timely intervention
Operational definition of spouse battering and child abuse
• Spousal battering is defined by physical assault, sexual coercion or injury, as measured by the revised Conflict Tactics Scales (CTS2).
• Child physical maltreatment is defined by severe or very severe levels of physical assault, as measured by the Parent-Child Conflict Tactics Scale (CTSPC)
Definition of risk assessment• Risk assessment is the process of identifying
and studying hazards to reduce the probability of their occurrence. (Boer, 1997)
• The process of evaluating individuals to (1) characterize the risk that they will commit violence in the future, and (2) develop interventions to manage or reduce that risk. (Monahan, 1994 )
Risk factors
Static risk factors
ImpactViolence
Dynamic risk factors
Physical Sexual
NeglectPsychological
Health
Mental Health
Risk assessment framework
Risk factors
Static risk factors
ImpactViolence
Dynamic risk factors
Physical Sexual
NeglectPsychological
Health
Mental Health
Prediction of risk
Development of risk assessment tools
• Phase A: Development of item pool– In-depth interviews and focus group
discussions with service users. – SWD and the Advisory Group on the Study on
Child Abuse and Spouse Battering– Pre-tested
• Phase B: Generation of Risk Assessment Tool – Representative household survey (December
2003 to August 2004) - A total of 5,049 adult respondents were successfully interviewed. The overall response rate achieved was 71%.
• Phase C: Preliminary logistic regression analysis – over 30 risk factors were identified for: – the perpetrator of spouse battering– the victim of spouse battering – the perpetrator of child abuse
• All these factors are significantly correlated to spouse battering and child abuse.
• Phase D: Factors included in the final model– Factors selected in the reduced model, that is,
the model used to design the risk assessment tools
Chinese Risk Assessment Tool for Spouse Battering Perpetrator
(Form A)
Logistic regression analysis (for perpetrators of spouse battering)
.000.000165.3691.056-8.540Constant
2.4851.7372.078.000164.180.091.731Partner’s disturbance (x13)
4.3431.9092.880.000125.457.2101.058Childhood witnessed parental violence (x12)
3.5441.7242.472.000124.264.184.905Criminal History (x11)
2.3401.3001.744.000113.742.150.556Face (x10)
.830.384.564.00418.428.197-.572Anger Management (x9)
1.5971.0541.297.01416.033.106.260Shifting Responsibility (x8)
2.3341.1671.651.00518.027.177.501Negative Attribution (x7)
2.2661.2721.698.000112.886.147.529Jealousy (x6)
2.8811.1221.798.01515.954.240.587Dominance (x5)
2.9541.0901.794.02215.284.254.585In-law Conflict (x4)
2.6171.1251.716.01216.284.215.540Indebtedness (x3)
3.6871.0261.945.04114.159.326.665Unemployment (x2)
3.3461.0681.890.02914.779.291.637Wife pregnancy/ adoption/postnatal
(within 1 year) (x1)
UpperLower
95% C.I. for Exp (B)Exp (B)Sig.dfWaldS.E.BRisk factor
For perpetrators of spouse battering, the required model equation is:
• A = -8.540 + 0.637X1 + 0.665X2 + 0.540X3+ 0.585X4 + 0.587X5 + 0.529X6 + 0.501X7+ 0.260X8 – 0.572X9 + 0.556X10 + 0.905X11+ 1.058X12+ 0.731X13
• P (risk (A)) = exp (A) / (1 + exp (A))
-2 log likelihood statistics for the logistic regression model
.000159.202-971.170Partner’s disturbance
.000122.783-952.960Childhood witnessed parental violence
.000122.022-952.580Criminal History
.000113.799-948.468Face
.00418.446-945.792Anger Management
.01416.100-944.619Shifting Responsibility
.00517.993-945.566Negative Attribution
.000112.765-947.951Jealousy
.01515.975-944.556Dominance
.03514.447-943.792In-law Conflict
.01615.831-944.484Indebtedness
.02814.832-943.985Unemployment
.03814.322-943.730Wife pregnancy/adoption/postnatal (within 1 year)
Sig. of the Chang
e
dfChange in -2 Log
Likelihood
Model Log Likelih
ood
Risk Factor
*The H-L test shows that the model explains the data well.
.61386.309Hosmer-Lemeshow (H-L) test for all risk factors regression analysis (Perpetrators)
Sig.Degree of freedom
Chi-square
100%33.07%66.93%Total
8.08%5.53%2.55%Happened
91.92%27.54%64.38%Not Happened
happenedNot Happened
TotalPredictedActual
All risk factors logistic regression analysis (perpetrators)
Classification table
• Sensitivity, which is the percentage of occurrences correctly predicted and is equal to (5.53%)/(8.08%) or 68.4%;
• Specificity, which is the percentage non-occurrences correctly predicted and is equal to (64.38)/(91.92%) or 70.0% ;
• Positive predictive value, which is the percentage of predicted occurrences that are correct and is equal to (5.53%)/(33.07%) or 16.7% ;
• Negative predictive value which is the percentage of predicted non-occurrences that are correct and is equal to (64.38%)/(66.93%) or 96.2% ;
• Overall accuracy, which is the percentage of predicted occurrences and non-occurrences that are correct and is equal to (64.38%+5.53%) or 69.9%.
• The optimal cut-off probability should be in the region of 7%.
0.0%
20.0%
40.0%
60.0%
80.0%
100.0%
95 85 75 65 55 45 35 25 15 9 7 5
Sensitivity /Specificity
SensitivitySpecificity
A Receiver Operating Characteristic (ROC) curve = 0.77
Chinese Risk Assessment Tool for Spouse Battering Victim
(Form B)
Logistic regression analysis (for victims of spouse battering)
.002.000149.839.849-5.996Constant
2.0551.3271.651.000120.184.112.502Feeling unsafe (x8)
2.3561.5701.924.000139.976.103.654Partner’s disturbance (x7)
4.7082.0083.075.000126.721.2171.123Childhood witnessed parental violence (x6)
5.5101.4562.832.00219.403.3401.041Sex Abuse History (x5)
3.1041.4432.116.000114.711.195.750Criminal History (x4)
.790.358.532.00219.800.202-.632Anger Management (x3)
2.9461.4222.047.000114.868.186.716Negative Attribution (x2)
2.7791.5132.051.000121.451.155.718Jealousy (x1)UpperLower
95% C.I. for Exp (B)Exp (B)Sig.dfWaldS.E.BRisk factor
For victims of spouse battering, the required model equation is:
• V = -5.996 + 0.718X1 + 0.716X2 - 0.632X3+ 0.750X4 + 1.041X5 + 1.123X6 + 0.654X7+ 0.502X8
• P (risk (v)) = exp (V) / (1 + exp (V))
-2 log likelihood statistics for the logistic regression model
.000119.329-855.471Feeling unsafe
.000136.877-864.245Partner’s disturbance
.000123.396-857.505Childhood witnessed parental violence
.00418.416-850.014Sex Abuse History
.000113.476-852.544Criminal History
.00219.834-850.724Anger Management
.000114.767-853.190Negative Attribution
.000121.270-856.442Jealousy
Sig. of the Change
dfChange in -2 Log Likelihood
Model Log Likelihood
Risk Factor
*The H-L test shows that the model explains the data well.
.055 815.211 Hosmer-Lemeshow (H-L) test for all risk factors regression analysis (Victims)
Sig.Degree of freedom
Chi-square
100% 33.02% 66.97% Total
7.10% 5.00% 2.10% Happened
92.89% 28.02% 64.87% Not Happened
happenedNot Happened
TotalPredictedActual
All risk factors logistic regression analysis (victims)
Classification table
• Sensitivity, which is the percentage of occurrences correctly predicted and is equal to (5%)/(7.10%) or 70.4%;
• Specificity, which is the percentage non-occurrences correctly predicted and is equal to (64.87%)/(92.89%) or 69.8% ;
• Positive predictive value, which is the percentage of predicted occurrences that are correct and is equal to (5%)/(33.02%) or 15.1% ;
• Negative predictive value which is the percentage of predicted non-occurrences that are correct and is equal to (64.87%)/(66.97%) or 96.9% ;
• Overall accuracy, which is the percentage of predicted occurrences and non-occurrences that are correct and is equal to (64.87%+5%) or 69.9%.
• The optimal cut-off probability should be in the region of 5.5%.
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
100.0%
95 85 75 65 55 45 35 25 15 9 7 5Sensitivity / Specificity
Sensitivity
Specificity
A Receiver Operating Characteristic (ROC) curve = 0.7687
Chinese Risk Assessment Tool for Child Abuse Perpetrator
(Form C)
Logistic regression analysis (for perpetrators of child abuse)
.000.000127.0641.567-8.150Constant
6.9602.6514.296.000135.048.2461.458Criminal History (x7)
5.0911.3682.639.00418.385.335.971Violence Approval (x6)
.771.233.424.00517.894.305-.858Anger Management (x5)
4.8371.9043.034.000121.775.2381.110Jealousy (x4)
2.8951.2761.922.00219.776.209.653Extended Family Influence (x3)
6.4412.1143.690.000121.098.2841.306Receiving CSSA (x2)
6.5661.0252.595.04414.052.474.953Unemployment (x1)
UpperLower
95% C.I. for Exp (B)Exp (B)
Sig.dfWaldS.E.BRisk factor
For perpetrators of spouse battering, the required model equation is:
• C = -8.150 + 0.953X1 + 1.306X2 + 0.653X3+ 1.110X4 – 0.858X5 + 0.971X6 + 1.458X7
• P (risk (c)) = exp (C) / (1 + exp (C))
-2 log likelihood statistics for the logistic regression model
.000130.226-400.222Criminal History
.00318.848-389.533Violence Approval
.00517.889-389.054Anger Management
.000121.821-396.019Jealousy
.001110.237-390.228Extended Family Influence
.000118.265-394.242Receiving CSSA
.02714.895-387.557Unemployment
Sig. of the Change
dfChange in -2 Log Likelihood
Model Log LikelihoodRisk Factor
*The H-L test shows that the model explains the data well.
.461 87.719 Hosmer-Lemeshow (H-L) test for all risk factors regression analysis (Perpetrators of child abuse)
Sig.Degree of freedom
Chi-square
100%30.00%70.00%Total
5.95%4.10%1.85%Happened
94.05%25.90%68.15%Not Happened
happenedNot Happened
TotalPredictedActual
All risk factors logistic regression analysis (perpetrators)
Classification table
• Sensitivity, which is the percentage of occurrences correctly predicted and is equal to (4.10%)/(5.95%) or 68.9%;
• Specificity, which is the percentage non-occurrences correctly predicted and is equal to (68.15%)/(94.05%) or 72.5% ;
• Positive predictive value, which is the percentage of predicted occurrences that are correct and is equal to (4.10%)/(30%) or 13.7% ;
• Negative predictive value which is the percentage of predicted non-occurrences that are correct and is equal to (68.15%)/(70%) or 97.4% ;
• Overall accuracy, which is the percentage of predicted occurrences and non-occurrences that are correct and is equal to (68.15%+4.1%) or 72.3%.
• The optimal cut-off probability should be in the region of 5.5%.
0.0%10.0%20.0%30.0%40.0%50.0%60.0%70.0%80.0%90.0%
100.0%
95 85 75 65 55 45 35 25 15 9 7 5Sensitivity /Specificity
Sensitivity
Specificity
A Receiver Operating Characteristic (ROC) curve = 0.7728
Usage of the tools
- The tool is designed to function as a triage, not for screening which could be done by using tools like AAS or CTS2.
- It is designed to help assessor assessing its probability of the occurrence of violence when risk factors continue to function and so decide the most appropriate way to handle it.
Target users of the tools
• The risk assessment tools are primarily designed for social workers, counselors and psychologists who have direct contact with the perpetrators and/or victims of domestic violence.
Functions of the risk assessment tools
• To facilitate early identification of domestic violence;
• To assist assessor in collecting fundamental information required to formulate further clinical assessment
Limitations of the risk assessment tools• Factors included in the tool reflect the most statistically
significant ones as demonstrated by the norm of respondents. Some factors with less significance may be valid for some particular cases.
• Accuracy of data collected depends heavily on the recollection of the respondents.
• While the tool is designed for self-report, there is always a possibility that the perpetrator may minimize, rationalize or deny acts of aggression against a spouse when responding to the questions asked in the tool.
• The scores generated by the tool show the optimum balance between sensitivity and specificity being calculated with statistical means.
• The scores are based on the assessment of risk factors. In addition, the scores only measure the probability of occurrence instead of the severity and frequency of violence. Direct assessment on the types, severity and frequency of violence used should be conducted.
• The tools should be treated as preliminary risk indicators; the assessor’s practice wisdom will be needed for final judgment. Second opinion from supervisor and senior practitioners should be sought.
People qualified to conduct risk assessment• social workers, counselors and psychologists
who have direct contact with the perpetrators and/or victims of domestic violence.
• People who wish to use the tools should ensure that the agency they work for has access to the information and resources needed to conduct a risk assessment of potential clients.
• To qualify to use the tools as part of the assessment procedure, the assessor should receive training in the tools’ usage.
• The training should enhance the assessor’s knowledge regarding the strengths and limitations of the tools and the standard procedures that need to be followed in conducting a risk assessment.
• An assessor with no prior experience of handling domestic violence should also receive trainingin understanding the dynamics of domestic violence, gender based violence and ways to elicit maximal information for objective judgment.
Administration of the tools
– Risk assessment should be conducted whenever the assessor can get in touch with a client.
– It is not necessary to wait for the appearance of physical signs like bruises and physical injuries, or the evolving of suicidal ideation before one is eligible for assessment.
– Only after the risk assessment can the assessor judges the potential risk the client bears.
– risk assessment should be done on a regular basis to monitor any changes in risk and to allow the assessor to readjust intervention to meet the client’s needs.
Eligible targets:• Perpetrator of spouse battering –
– people who reported or being complained of using violence against partner, usually the primary aggressor in cases of mutual combat.
• Victim of spouse battering –– people who reported being abused by a partner, showing fear
towards partner or being stalked by partner, usually the primary victim in cases of mutual combat.
• Perpetrator of child abuse –– people who reported or being complained of using violence
against a child, or neglect the needs for healthy development of a child.
– In case of mutual combat, • the assessor should identify the primary
aggressor by looking at the types and frequency of violence used, the severity of harm inflicted on the other partner, fear induced, power and control issues.
• If the assessor still finds it difficult to differentiate, the client may be asked to complete two sets of the risk assessment tools, one for the perpetrator of spouse battering and one for the victim.
– Non-perpetrator or non-victim may be assessed if assessor finds it necessary.
• Clients may display behaviors related to risk factors, for instance, unemployment and in-law conflict.
Use of data in clinical risk assessment
(1) consider all risk factors supported in the study
In-law Conflict
Extended Family Influence
Indebtedness
Receiving CSSA
Income
Unemployment
Wife pregnancy/ adoption/postnatal
Disability
Chronic ill
Perpetrator(Form C)
Victim (Form B)
Perpetrator(Form A)
Child abuseSpouse battering
Stressful Conditions
Social Desirability
Depressive Symptoms
Violence Approval
Substance Abuse
Anger Management
Shifting Responsibility
Negative Attribution
Jealousy
Dominance
Relationship Distress
Feeling unsafe
Afraid of partner
Partner’s disturbance
Childhood witnessed parental violence
Child Neglect
Sex Abuse History
Criminal History
Suicidal Ideation
Social Support
Self-esteem
Face
(2) The risk assessment tools are designed to measure the probability of occurrence of spouse battering and child abuse by detecting the presence of various risk factors that have been found to significantly correlate with the occurrence of domestic violence.
(3) Although the assessment is valid for the time of the interview, it does not measure changes in factors over time. The client should be reassessed on a regular basis to monitor changes he or she may demonstrate.
(4) employ multiple sources of information; the more sources of information the better
(5) be victim-informed(6) Instrument improves clinical judgment – but
clinician wisdom also plays important role(7) Risk assessment should lead to risk management.
• The tool is designed to function as a triage. it is designed to help assessor evaluate the urgency of a case by assessing its probability of the occurrence of violence when risk factors continue to function and so decide the most appropriate way to handle
• The more factors that are detected, the higher the likelihood domestic violence will occur.
Reference• Boer, D. P., Hart, S. D., Kropp, P. R., & Webster, C. D. (1997). Manual for
the Sexual Violence Risk - 20. British Columbia: The British Columbia Institute Against Family Violence.
• Kantor, G. K., & Jasinski, J. L. (1998). Dynamics and risk factors in partner violence. In J. L. Jasinski & L. M. Williams (Eds.), Partner violence: A comprehensive review of 20 years of research . USA: Sage.
• Monahan, J., & Steadman, H. J. (1996). Violent storms and violent people: How meteorology can inform risk communication in mental health law. American Psychologist, 51(9), 931-938.
• Mrazek, P. J., & Haggerty, R. J. (1994). Reducing risks for mental disorders: Frontiers for preventive intervention. Washington, DC: National Academy Press.
• Vincent C. (2001)Clinical Risk Management – Enhancing patient safety, BMJ, (2nd Edition)
• Kipp D. J. and Loflin E. M. (1996), Emergency Incident Risk Management, VNR.
The End
~ Thank You ~